Rewrite of fixed_compute_best_predictor functions

The code of fixed_compute_best_predictor was presumable once optimized,
but it appears a much more readable form is now equally fast or even faster,
see https://github.com/xiph/flac/pull/337 for details. 
1 file changed